6
$\begingroup$

I have time series data from mobile sensors for different motions such as walking, pushups, dumbellifts, rowing and so on. All these motions have different length of time series. For classifying them using Dynamic Time Warping (DTW), how do I choose an appropriate window size that will give good results?

$\endgroup$
1
$\begingroup$

If you have enough data, use cross validation.

If you don't have a lot of data, use cross validation on a similar dataset, and transfer the window size (the UCR archive has a bunch of similar dataset)

Don't forget, that the best warping window size depends on the amount of training data. As you get more data, you can have a smaller warping window, see fig 6 of http://www.cs.ucr.edu/~eamonn/DTW_myths.pdf

eamonn

| improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.